To run this code, you will need recent versions of R and Python. Our analysis was run with R 4.3.1 and Python 3.10.12. Python is used primarily for scraping basketball-reference.com, and depends on the requests, beautifulsoup4, and lxml packages. The bulk of the analysis is written in R and depends on the following R packages:
data.table,rio(data i/o and wrangling)glue(string formatting model specification)lubridate(handling dates)sandwich,miceadds(logit with robust standard errors)margins(average marginal effects)modelsummary(nice model output)flextable,kableExtra(.docx and .xlsx output)
If you have Nix installed, you can create the exact environment used in our analysis with the included default.nix. Just run nix-build from the root directory to install all necessary dependencies (this may take a while), and nix-shell for an interactive shell.
The replication relies on the Stata data file andata.dta produced from Biegert, Kühhirt & Van Lancker's replication package. The code and base data are available at https://doi.org/10.17605/osf.io/ntwdy. We include andata.dta created using Stata 17.0 BE based on the replication package as downloaded in August of 2023.
The full code for analysis consists of six files:
01_replication_data_to_csv.R: Convert BKVL's .dta to a .csv.02_scrape_br.py: Scrape game logs, player data, and season data from basketball-reference.com. Note: this script makes close to 15,000 requests to basketball-reference.com. With the necessary rate limiting, it will take up to 24 hours to run. By default the script caches its data, which amounts to approximately 4.6GB uncompressed.03_process_and_merge.R: Process game logs and player data, and merge into a single dataset at the player–season level.04_replication.R: Replicate BKVL's analysis using their data and corrected data with comparisons.05_expanded_analysis.R: Our main analyses.06_forest_plot.R: Create the summary forest plot comparing models.
With the dependencies (listed above) installed the full analysis can be run with the following shell commands. (Note that the second script, 02_scrape_br.py is quite time- and disk-intensive. It will take close to 24 hours to run and will download approximately 4.6GB of data to your computer.)
Rscript scripts/01_replication_data_to_csv.R
python scripts/02_scrape_br.py
Rscript scripts/03_process_and_merge.R
Rscript scripts/04_replication.R
Rscript scripts/05_expanded_analysis.R
Rscript scripts/06_forest_plot.R
Processed data are stored in the data directory, saved model-fit objects are stored in the models directory, and tables and figures are stored in the output directory.
This work is licensed under CC BY 4.0
